gem-llm-supervisor

To start dual vLLM instances with a gateway and admin UI, the supervisor enforces model readiness by polling with a timeout before bringing up dependent services in the correct order.

A consolidated health check verifies local services and the external Cloudflare tunnel endpoint by running port-specific checks and providing deterministic status logs across the full stack.

vLLM stack startup failures after a reboot often stem from GPU PID leftovers, occupied ports, or a down Cloudflare tunnel, which the supervisor addresses through automated recovery and troubleshooting playbooks.

Troubleshooting a multi-service LLM deployment involves checking consolidated health status, identifying service-specific errors, and inspecting port availability to pinpoint the exact failure point.

Yes, you can restart the full stack automatically by executing the supervisor command, which handles stopping existing processes and bringing the dual vLLM, gateway, and admin UI back up sequentially.

You need a supervisor for integrated orchestration when managing multi-service deployments, ensuring correct startup sequencing, and diagnosing reachability issues across local services and external tunnels.